diff --git a/docs/_static/css/tvm_theme.css b/docs/_static/css/tvm_theme.css
new file mode 100644
index 000000000000..5e0838abf6cb
--- /dev/null
+++ b/docs/_static/css/tvm_theme.css
@@ -0,0 +1,11 @@
+.rst-content .hidden-section {
+    display: none;
+}
+
+.rst-toc .hidden-section {
+    display: none;
+}
+
+nav .hidden-section {
+    display: inherit;
+}
diff --git a/docs/api/python/build.rst b/docs/api/python/build.rst
new file mode 100644
index 000000000000..4cf936910e1b
--- /dev/null
+++ b/docs/api/python/build.rst
@@ -0,0 +1,5 @@
+tvm.build
+---------
+.. autofunction:: tvm.lower
+
+.. autofunction:: tvm.build
diff --git a/docs/api/python/collection.rst b/docs/api/python/collection.rst
new file mode 100644
index 000000000000..1eb83861d940
--- /dev/null
+++ b/docs/api/python/collection.rst
@@ -0,0 +1,6 @@
+tvm.collections
+---------------
+Collections contains data structures used in TVM DSL.
+
+.. automodule:: tvm.collections
+    :members:
diff --git a/docs/api/python/dev.rst b/docs/api/python/dev.rst
new file mode 100644
index 000000000000..11dcd9609ccb
--- /dev/null
+++ b/docs/api/python/dev.rst
@@ -0,0 +1,65 @@
+Developer API
+-------------
+This page contains modules that are used by developers of TVM.
+
+tvm.node
+~~~~~~~~
+Node is the base class of all TVM AST. Normally user do not
+need to touch this api.
+
+.. autoclass:: tvm.node.NodeBase
+    :members:
+
+.. autoclass:: tvm.node.Node
+    :members:
+
+.. autofunction:: tvm.register_node
+
+tvm.expr
+~~~~~~~~
+.. automodule:: tvm.expr
+   :members:
+   :undoc-members:
+
+tvm.codegen
+~~~~~~~~~~~
+.. automodule:: tvm.codegen
+   :members:
+   :undoc-members:
+
+tvm.stmt
+~~~~~~~~
+.. automodule:: tvm.stmt
+   :members:
+   :undoc-members:
+
+tvm.ir_pass
+~~~~~~~~~~~
+.. automodule:: tvm.ir_pass
+   :members:
+
+.. autosummary::
+
+   tvm.ir_pass.Inline
+   tvm.ir_pass.Simplify
+   tvm.ir_pass.ConvertSSA
+   tvm.ir_pass.VerifySSA
+   tvm.ir_pass.CanonicalSimplify
+   tvm.ir_pass.StorageFlatten
+   tvm.ir_pass.VectorizeLoop
+   tvm.ir_pass.UnrollLoop
+   tvm.ir_pass.StorageSync
+   tvm.ir_pass.MakeAPI
+   tvm.ir_pass.SplitHostDevice
+   tvm.ir_pass.InjectVirtualThread
+   tvm.ir_pass.LoopPartition
+   tvm.ir_pass.RemoveNoOp
+   tvm.ir_pass.SplitPipeline
+   tvm.ir_pass.LowerThreadAllreduce
+   tvm.ir_pass.NarrowChannelAccess
+
+
+tvm.make
+~~~~~~~~
+.. automodule:: tvm.make
+   :members:
diff --git a/docs/api/python/function.rst b/docs/api/python/function.rst
new file mode 100644
index 000000000000..50a79729bc9a
--- /dev/null
+++ b/docs/api/python/function.rst
@@ -0,0 +1,7 @@
+tvm.Function
+------------
+.. autoclass:: tvm.Function
+
+.. autofunction:: tvm.register_func
+
+.. autofunction:: tvm.get_global_func
diff --git a/docs/api/python/index.rst b/docs/api/python/index.rst
new file mode 100644
index 000000000000..c359f9570d70
--- /dev/null
+++ b/docs/api/python/index.rst
@@ -0,0 +1,16 @@
+Python API
+==========
+
+.. toctree::
+   :maxdepth: 2
+
+   tvm
+   tensor
+   schedule
+   build
+   module
+   ndarray
+   collection
+   node
+   function
+   dev
diff --git a/docs/api/python/module.rst b/docs/api/python/module.rst
new file mode 100644
index 000000000000..112a92e367ab
--- /dev/null
+++ b/docs/api/python/module.rst
@@ -0,0 +1,9 @@
+tvm.module
+----------
+.. autoclass:: tvm.module.Module
+    :members:
+    :inherited-members:
+
+.. autofunction:: tvm.module.load
+
+.. autofunction:: tvm.module.enabled
diff --git a/docs/api/python/ndarray.rst b/docs/api/python/ndarray.rst
new file mode 100644
index 000000000000..f636860946ba
--- /dev/null
+++ b/docs/api/python/ndarray.rst
@@ -0,0 +1,16 @@
+tvm.ndarray
+-----------
+tvm.ndarray provides a minimum runtime array API to testing out
+the correctness of the program.
+
+.. autoclass:: tvm.ndarray.TVMContext
+    :members:
+
+.. autoclass:: tvm.ndarray.NDArray
+    :members:
+    :inherited-members:
+
+.. autofunction:: tvm.cpu
+.. autofunction:: tvm.gpu
+.. autofunction:: tvm.opencl
+.. autofunction:: tvm.ndarray.array
diff --git a/docs/api/python/schedule.rst b/docs/api/python/schedule.rst
new file mode 100644
index 000000000000..4596ba62b997
--- /dev/null
+++ b/docs/api/python/schedule.rst
@@ -0,0 +1,17 @@
+tvm.schedule
+------------
+The `tvm.schedule` module contains classes of scheduling
+structure of tvm.
+
+.. autoclass:: tvm.schedule.IterVar
+    :members:
+
+.. autofunction:: tvm.create_schedule
+
+.. autoclass:: tvm.schedule.Schedule
+    :members:
+    :inherited-members:
+
+.. autoclass:: tvm.schedule.Stage
+    :members:
+    :inherited-members:
diff --git a/docs/api/python/tensor.rst b/docs/api/python/tensor.rst
new file mode 100644
index 000000000000..b7c2cd10baba
--- /dev/null
+++ b/docs/api/python/tensor.rst
@@ -0,0 +1,28 @@
+tvm.tensor
+----------
+The `tvm.tensor` module contains declaration of Tensor
+and Operation class for computation declaration.
+
+.. autoclass:: tvm.tensor.Tensor
+    :members:
+    :inherited-members:
+
+.. autoclass:: tvm.tensor.Operation
+    :members:
+    :inherited-members:
+
+.. autoclass:: tvm.tensor.ComputeOp
+    :members:
+    :show-inheritance:
+
+.. autoclass:: tvm.tensor.PlaceholderOp
+    :members:
+    :show-inheritance:
+
+.. autoclass:: tvm.tensor.ScanOp
+    :members:
+    :show-inheritance:
+
+.. autoclass:: tvm.tensor.ExternOp
+    :members:
+    :show-inheritance:
diff --git a/docs/api/python/tvm.rst b/docs/api/python/tvm.rst
new file mode 100644
index 000000000000..b16911aa18de
--- /dev/null
+++ b/docs/api/python/tvm.rst
@@ -0,0 +1,40 @@
+tvm
+---
+tvm is a library root namespace contains functions for
+declaring computation.
+
+.. autosummary::
+
+   tvm.load_json
+   tvm.save_json
+   tvm.var
+   tvm.const
+   tvm.convert
+   tvm.placeholder
+   tvm.compute
+   tvm.scan
+   tvm.extern
+   tvm.call_packed
+   tvm.decl_buffer
+   tvm.reduce_axis
+   tvm.thread_axis
+   tvm.sum
+   tvm.min
+   tvm.max
+
+.. autofunction:: tvm.load_json
+.. autofunction:: tvm.save_json
+.. autofunction:: tvm.var
+.. autofunction:: tvm.const
+.. autofunction:: tvm.convert
+.. autofunction:: tvm.placeholder
+.. autofunction:: tvm.compute
+.. autofunction:: tvm.scan
+.. autofunction:: tvm.extern
+.. autofunction:: tvm.call_packed
+.. autofunction:: tvm.decl_buffer
+.. autofunction:: tvm.reduce_axis
+.. autofunction:: tvm.thread_axis
+.. autofunction:: tvm.sum
+.. autofunction:: tvm.min
+.. autofunction:: tvm.max
diff --git a/docs/conf.py b/docs/conf.py
index e887163f3e42..e97033d4960c 100644
--- a/docs/conf.py
+++ b/docs/conf.py
@@ -26,7 +26,6 @@
 libpath = os.path.join(curr_path, '../python/')
 sys.path.insert(0, libpath)
 
-
 # -- General configuration ------------------------------------------------
 
 # General information about the project.
@@ -51,11 +50,15 @@
 extensions = [
     'sphinx.ext.autodoc',
     'sphinx.ext.autosummary',
+    'sphinx.ext.intersphinx',
     'sphinx.ext.napoleon',
     'sphinx.ext.mathjax',
     'sphinx_gallery.gen_gallery',
 ]
 
+breathe_projects = {'tvm' : 'doxygen/xml/'}
+breathe_default_project = 'tvm'
+
 # Add any paths that contain templates here, relative to this directory.
 templates_path = ['_templates']
 
@@ -132,7 +135,7 @@
 # Add any paths that contain custom static files (such as style sheets) here,
 # relative to this directory. They are copied after the builtin static files,
 # so a file named "default.css" will overwrite the builtin "default.css".
-#html_static_path = ['_static']
+html_static_path = ['_static']
 
 # Output file base name for HTML help builder.
 htmlhelp_basename = project + 'doc'
@@ -153,12 +156,19 @@
 def run_doxygen(folder):
     """Run the doxygen make command in the designated folder."""
     try:
-        retcode = subprocess.call("cd %s; make doxygen" % folder, shell=True)
+        retcode = subprocess.call("cd %s; make doc" % folder, shell=True)
         if retcode < 0:
             sys.stderr.write("doxygen terminated by signal %s" % (-retcode))
     except OSError as e:
         sys.stderr.write("doxygen execution failed: %s" % e)
 
+intersphinx_mapping = {
+    'python': ('https://docs.python.org/{.major}'.format(sys.version_info), None),
+    'numpy': ('http://docs.scipy.org/doc/numpy/', None),
+    'scipy': ('http://docs.scipy.org/doc/scipy/reference', None),
+    'matplotlib': ('http://matplotlib.org/', None),
+}
+
 examples_dirs = ['../tutorials/python']
 gallery_dirs = ['tutorials']
 
@@ -171,18 +181,24 @@ def generate_doxygen_xml(app):
 def setup(app):
     # Add hook for building doxygen xml when needed
     # no c++ API for now
-    # app.connect("builder-inited", generate_doxygen_xml)
+    app.connect("builder-inited", generate_doxygen_xml)
+    app.add_stylesheet('css/tvm_theme.css')
     app.add_config_value('recommonmark_config', {
             'url_resolver': lambda url: github_doc_root + url,
             }, True)
     app.add_transform(AutoStructify)
 
+
 sphinx_gallery_conf = {
     'backreferences_dir': 'gen_modules/backreferences',
-    'doc_module': ('tvm'),
-    'reference_url': {
-        'tvm': None
-    },
+    'doc_module': ('tvm', 'numpy'),
+'reference_url': {
+    'tvm': None,
+    'matplotlib': 'http://matplotlib.org',
+    'numpy': 'http://docs.scipy.org/doc/numpy-1.9.1'},
     'examples_dirs': examples_dirs,
-    'gallery_dirs': gallery_dirs
+    'gallery_dirs': gallery_dirs,
+    'find_mayavi_figures': False,
+    'filename_pattern': '.py',
+    'expected_failing_examples': []
 }
diff --git a/docs/index.rst b/docs/index.rst
index e025888d0f53..f265044b832b 100644
--- a/docs/index.rst
+++ b/docs/index.rst
@@ -9,7 +9,7 @@ Contents
 .. toctree::
    :maxdepth: 1
 
-   how_to/contribute
    how_to/install
    tutorials/index
-   python/api
+   api/python/index
+   how_to/contribute
diff --git a/docs/python/api.rst b/docs/python/api.rst
deleted file mode 100644
index 017a6fc3bf92..000000000000
--- a/docs/python/api.rst
+++ /dev/null
@@ -1,108 +0,0 @@
-Python API
-==========
-
-tvm
----
-tvm is a library root namespace contains functions for
-declaring computation.
-
-.. autofunction:: tvm.load_json
-
-.. autofunction:: tvm.save_json
-
-.. autofunction:: tvm.var
-
-.. autofunction:: tvm.convert
-
-.. autofunction:: tvm.placeholder
-
-.. autofunction:: tvm.compute
-
-.. autofunction:: tvm.scan
-
-.. autofunction:: tvm.extern
-
-.. autofunction:: tvm.reduce_axis
-
-.. autofunction:: tvm.sum
-
-tvm.tensor
-----------
-The `tvm.tensor` module contains declaration of Tensor
-and Operation class for computation declaration.
-
-.. autoclass:: tvm.tensor.Tensor
-    :members:
-    :inherited-members:
-
-.. autoclass:: tvm.tensor.Operation
-    :members:
-    :inherited-members:
-
-tvm.schedule
-------------
-.. autofunction:: tvm.create_schedule
-
-.. autoclass:: tvm.schedule.Schedule
-    :members:
-
-.. autoclass:: tvm.schedule.Stage
-    :members:
-
-tvm.build
----------
-
-.. autofunction:: tvm.lower
-.. autofunction:: tvm.build
-
-tvm.ndarray
------------
-tvm.ndarray provides a minimum runtime array API to testing out
-the correctness of the program.
-
-.. autofunction:: tvm.cpu
-.. autofunction:: tvm.gpu
-.. autofunction:: tvm.vpi
-.. autofunction:: tvm.opencl
-.. autofunction:: tvm.ndarray.array
-
-.. autoclass:: tvm.ndarray.TVMContext
-    :members:
-
-.. autoclass:: tvm.ndarray.NDArray
-    :members:
-    :inherited-members:
-
-tvm.Function
-------------
-
-.. autofunction:: tvm.register_func
-
-.. autoclass:: tvm.Function
-
-tvm.module
-----------
-.. autofunction:: tvm.module.load
-
-.. autofunction:: tvm.module.load
-
-.. autoclass:: tvm.module.Module
-    :members:
-    :inherited-members:
-
-tvm.node
---------
-tvm.node provides
-
-.. autofunction:: tvm.register_node
-
-.. autoclass:: tvm.node.NodeBase
-    :members:
-
-.. autoclass:: tvm.node.Node
-    :members:
-
-tvm.expr
---------
-.. automodule:: tvm.expr
-    :members:
diff --git a/python/tvm/_ctypes/_function.py b/python/tvm/_ctypes/_function.py
index d5898be550c0..108136802ce9 100644
--- a/python/tvm/_ctypes/_function.py
+++ b/python/tvm/_ctypes/_function.py
@@ -136,7 +136,27 @@ def _make_tvm_args(args, temp_args):
 
 
 class Function(object):
-    """The Function object used in TVM.
+    """The PackedFunc object used in TVM.
+
+    Function plays an key role to bridge front and backend in TVM.
+    Function provide a type-erased interface, you can call function with positional arguments.
+
+    The compiled module returns Function.
+    TVM backend also registers and exposes its API as Functions.
+    For example, the developer function exposed in tvm.ir_pass are actually
+    C++ functions that are registered as PackedFunc
+
+    The following are list of common usage scenario of tvm.Function.
+
+    - Automatic exposure of C++ API into python
+    - To call PackedFunc from python side
+    - To call python callbacks to inspect results in generated code
+    - Bring python hook into C++ backend
+
+    See Also
+    --------
+    tvm.register_func: How to register global function.
+    tvm.get_global_func: How to get global function.
     """
     __slots__ = ["handle", "is_global"]
     # pylint: disable=no-member
@@ -299,6 +319,26 @@ def register_func(func_name, f=None):
     -------
     fregister : function
         Register function if f is not specified.
+
+    Examples
+    --------
+    The following code registers my_packed_func as global function.
+    Note that we simply get it back from global function table to invoke
+    it from python side. However, we can also invoke the same function
+    from C++ backend, or in the compiled TVM code.
+
+    .. code-block:: python
+
+      targs = (10, 10.0, "hello")
+      @tvm.register_func
+      def my_packed_func(*args):
+          assert(tuple(args) == targs)
+          return 10
+      # Get it out from global function table
+      f = tvm.get_global_func("my_packed_func")
+      assert isinstance(f, tvm.nd.Function)
+      y = f(*targs)
+      assert y == 10
     """
     if callable(func_name):
         f = func_name
@@ -328,7 +368,7 @@ def get_global_func(name):
 
     Returns
     -------
-    func : tvm.nd.Function
+    func : tvm.Function
         The function to be returned.
     """
     handle = FunctionHandle()
@@ -355,27 +395,60 @@ def list_global_func_names():
     return fnames
 
 
-def _init_api_functions(root_namespace):
-    """List and add all the functions to current module."""
-    module_obj = sys.modules["%s.api" % root_namespace]
-    module_internal = sys.modules["%s._api_internal" % root_namespace]
+def _get_api(f):
+    flocal = f
+    def my_api_func(*args):
+        """
+
+        This is a type erased API that calls into Global PackedFunc.
+        These APIs corresponds to functions registered from C++ backend
+        and can be used as developer functions.
+
+        args : list
+          The positional arguments to the function call.
+
+        Returns
+        -------
+        value : int, float, None, Node or Function
+        The result of the API function call.
+        """
+        return flocal(*args)
+    return my_api_func
+
+def _init_api(mod):
+    """Initialize api for a given module name
+
+    mod : str
+       The name of the module.
+    """
+    module = sys.modules[mod]
     namespace_match = {
-        "_make_": sys.modules["%s.make" % root_namespace],
-        "_arith_": sys.modules["%s.arith" % root_namespace],
-        "_pass_": sys.modules["%s.ir_pass" % root_namespace],
-        "_codegen_": sys.modules["%s.codegen" % root_namespace],
-        "_module_": sys.modules["%s.module" % root_namespace],
-        "_schedule_": sys.modules["%s.schedule" % root_namespace]
+        "_make_": "tvm.make",
+        "_arith_": "tvm.arith",
+        "_pass_": "tvm.ir_pass",
+        "_codegen_": "tvm.codegen",
+        "_module_": "tvm.module",
+        "_schedule_": "tvm.schedule"
     }
     for name in list_global_func_names():
         fname = name
-        target_module = module_internal if name.startswith('_') else module_obj
+        target = "tvm.api"
         for k, v in namespace_match.items():
             if name.startswith(k):
                 fname = name[len(k):]
-                target_module = v
+                target = v
+        if target != mod:
+            continue
+        if mod == "tvm.api" and name.startswith("_"):
+            target_module = sys.modules["tvm._api_internal"]
+        else:
+            target_module = module
+
         f = get_global_func(name)
-        setattr(target_module, fname, f)
+        ff = _get_api(f)
+        ff.__name__ = fname
+        ff.__doc__ = ("TVM PackedFunc %s. " % fname)
+        setattr(target_module, ff.__name__, ff)
 
 
 def _init_module_module(module_class):
diff --git a/python/tvm/_ctypes/_node.py b/python/tvm/_ctypes/_node.py
index 6f8d8905e1c5..29c4c1224a34 100644
--- a/python/tvm/_ctypes/_node.py
+++ b/python/tvm/_ctypes/_node.py
@@ -186,13 +186,14 @@ def register_node(type_key=None):
     type_key : str or cls
         The type key of the node
     """
+    node_name = type_key if isinstance(type_key, str) else type_key.__name__
+
+    def register(cls):
+        """internal register function"""
+        NODE_TYPE[node_name] = cls
+        return cls
+
     if isinstance(type_key, str):
-        def register(cls):
-            """internal register function"""
-            NODE_TYPE[type_key] = cls
-            return cls
         return register
     else:
-        cls = type_key
-        NODE_TYPE[cls.__name__] = cls
-        return cls
+        return register(type_key)
diff --git a/python/tvm/api.py b/python/tvm/api.py
index fffdd20bf86f..1f697558917b 100644
--- a/python/tvm/api.py
+++ b/python/tvm/api.py
@@ -8,13 +8,14 @@
 from ._ctypes._node import register_node, NodeBase
 from ._ctypes._node import convert_to_node as _convert_to_node
 from ._ctypes._function import Function
-from ._ctypes._function import _init_api_functions, register_func, get_global_func
+from ._ctypes._function import _init_api, register_func, get_global_func
 from ._ctypes._function import convert_to_tvm_func as _convert_tvm_func
 from . import _api_internal
 from . import _base
 from . import make as _make
 from . import expr as _expr
 from . import tensor as _tensor
+from . import schedule as _schedule
 from . import collections as _collections
 
 int32 = "int32"
@@ -31,6 +32,27 @@ def const(value, dtype=None):
     return _api_internal._const(value, dtype)
 
 
+def convert(value):
+    """Convert value to TVM node or function.
+
+    Parameters
+    ----------
+    value : python value
+
+    Returns
+    -------
+    tvm_val : Node or Function
+        Converted value in TVM
+    """
+    if isinstance(value, (Function, NodeBase)):
+        return value
+
+    if callable(value):
+        return _convert_tvm_func(value)
+    else:
+        return _convert_to_node(value)
+
+
 def load_json(json_str):
     """Load tvm object from json_str.
 
@@ -179,8 +201,8 @@ def scan(init, update, state_placeholder, inputs=None, name="scan"):
     .. code-block:: python
 
       # The following code is equivalent to numpy.cumsum
-      m = tvm.Var("m")
-      n = tvm.Var("n")
+      m = tvm.var("m")
+      n = tvm.var("n")
       X = tvm.placeholder((m, n), name="X")
       s_state = tvm.placeholder((m, n))
       s_init = tvm.compute((1, n), lambda _, i: X[0, i])
@@ -281,7 +303,10 @@ def decl_buffer(shape, dtype=None,
                 strides=None,
                 byte_offset=None,
                 offset_alignment=0):
-    """Decleare a new symbolic buffer
+    """Decleare a new symbolic buffer.
+
+    Normally buffer is created automatically during lower and build.
+    This is only needed if user want to specify their own buffer layout.
 
     Parameters
     ----------
@@ -370,6 +395,11 @@ def thread_axis(dom=None, tag='', name=''):
 
     name : str, optional
         The name of the var.
+
+    Returns
+    -------
+    axis : IterVar
+        The thread itervar.
     """
     if isinstance(dom, _base.string_types):
         tag, dom = dom, None
@@ -446,7 +476,7 @@ def min(lhs, rhs=None, axis=None, where=None):
     """
     if rhs and axis:
         raise ValueError("Can only take one argument, rhs or axis")
-    if isinstance(rhs, (_collections.IterVar, list)):
+    if isinstance(rhs, (_schedule.IterVar, list)):
         axis, rhs = rhs, axis
     if rhs:
         return _make.Min(lhs, rhs)
@@ -479,7 +509,7 @@ def max(lhs, rhs=None, axis=None, where=None):
     """
     if rhs and axis:
         raise ValueError("Can only take one argument, rhs or axis")
-    if isinstance(rhs, (_collections.IterVar, list)):
+    if isinstance(rhs, (_schedule.IterVar, list)):
         axis, rhs = rhs, axis
     if rhs:
         return _make.Max(lhs, rhs)
@@ -487,26 +517,4 @@ def max(lhs, rhs=None, axis=None, where=None):
     x = _make.Reduce("Max", expr, axis, where)
     return x
 
-
-def convert(value):
-    """Convert value to TVM node or function.
-
-    Parameters
-    ----------
-    value : python value
-
-    Returns
-    -------
-    tvm_val : Node or Function
-        Converted value in TVM
-    """
-    if isinstance(value, (Function, NodeBase)):
-        return value
-
-    if callable(value):
-        return _convert_tvm_func(value)
-    else:
-        return _convert_to_node(value)
-
-
-_init_api_functions("tvm")
+_init_api("tvm.api")
diff --git a/python/tvm/arith.py b/python/tvm/arith.py
index a8886fcf2150..ed086bfdebe5 100644
--- a/python/tvm/arith.py
+++ b/python/tvm/arith.py
@@ -2,6 +2,7 @@
 from __future__ import absolute_import as _abs
 
 from ._ctypes._node import NodeBase, register_node
+from ._ctypes._function import _init_api
 from . import _api_internal
 
 class IntSet(NodeBase):
@@ -36,3 +37,5 @@ class StrideSet(IntSet):
 class ModularSet(IntSet):
     """Represent range of (coeff * x + base) for x in Z """
     pass
+
+_init_api("tvm.arith")
diff --git a/python/tvm/build.py b/python/tvm/build.py
index 5f3f3298429c..2bd99a185d8f 100644
--- a/python/tvm/build.py
+++ b/python/tvm/build.py
@@ -134,10 +134,10 @@ def build(sch,
     fapi = ir_pass.LowerThreadAllreduce(fapi, warp_size)
     fsplits = [s for s in ir_pass.SplitHostDevice(fapi)]
     if len(fsplits) > 1:
-        mhost = codegen.build(fsplits[0], target_host)
+        mhost = codegen.build_module(fsplits[0], target_host)
         if target:
-            mdev = codegen.build(fsplits[1:], target)
+            mdev = codegen.build_module(fsplits[1:], target)
             mhost.import_module(mdev)
         return mhost
     else:
-        return codegen.build(fsplits[0], target)
+        return codegen.build_module(fsplits[0], target)
diff --git a/python/tvm/codegen.py b/python/tvm/codegen.py
index 02dda155c19a..fbb102d464e8 100644
--- a/python/tvm/codegen.py
+++ b/python/tvm/codegen.py
@@ -1 +1,39 @@
-"""Code generation related functions"""
+"""Code generation related functions."""
+from ._ctypes._function import _init_api
+
+def build_module(lowered_func, target):
+    """Build lowered_func into Module.
+
+    Parameters
+    ----------
+    lowered_func : LoweredFunc
+        The lowered function
+
+    target : str
+        The target module type.
+
+    Returns
+    -------
+    module : Module
+        The corressponding module.
+    """
+    return _Build(lowered_func, target)
+
+
+def enabled(target):
+    """Whether target is enabled for codegen.
+
+    Parameters
+    ----------
+    target : str
+        The target module type.
+
+    Returns
+    -------
+    enabled : boolean
+        Whether the target module is enabled.
+    """
+    return _Enabled(target)
+
+
+_init_api("tvm.codegen")
diff --git a/python/tvm/collections.py b/python/tvm/collections.py
index 40ea219faa46..c2af29661702 100644
--- a/python/tvm/collections.py
+++ b/python/tvm/collections.py
@@ -2,11 +2,16 @@
 from __future__ import absolute_import as _abs
 from ._ctypes._node import NodeBase, register_node
 from . import _api_internal
-from . import expr as _expr
 
 @register_node
 class Array(NodeBase):
-    """Array container of TVM"""
+    """Array container of TVM.
+
+    You do not need to create Array explicitly.
+    Normally python list and tuple will be converted automatically
+    to Array during tvm function call.
+    You may get Array in return values of TVM function call.
+    """
     def __getitem__(self, i):
         if isinstance(i, slice):
             start = i.start if i.start is not None else 0
@@ -26,7 +31,13 @@ def __repr__(self):
 
 @register_node
 class Map(NodeBase):
-    """Map container of TVM"""
+    """Map container of TVM.
+
+    You do not need to create Map explicitly.
+    Normally python dict will be converted automatically
+    to Array during tvm function call.
+    You may get Map in return values of TVM function call.
+    """
     def __getitem__(self, k):
         return _api_internal._MapGetItem(self, k)
 
@@ -47,22 +58,12 @@ def __repr__(self):
 
 @register_node
 class Range(NodeBase):
-    """Represent range in TVM"""
-    pass
-
-
-@register_node
-class IterVar(NodeBase, _expr.ExprOp):
-    """Represent iteration variable."""
-    DataPar = 0
-    ThreadIndex = 1
-    CommReduce = 2
-    Ordered = 3
-    DimInfo = 4
-    Unrolled = 5
-    Vectorized = 6
-    Parallelized = 7
+    """Represent range in TVM.
 
+    You do not need to create Range explicitly.
+    Python list and tuple will be converted automatically to Range in api functions.
+    """
+    pass
 
 @register_node
 class LoweredFunc(NodeBase):
diff --git a/python/tvm/expr.py b/python/tvm/expr.py
index a9e9750e8e42..28aa4c92079a 100644
--- a/python/tvm/expr.py
+++ b/python/tvm/expr.py
@@ -1,4 +1,19 @@
-"""Module to declare Expression class"""
+"""Expression AST Node in TVM.
+
+User do not need to deal with expression AST node directly.
+But they can be helpful for developer to do quick proptyping.
+While not displayed in the document and python file.
+Each expression node have subfields that can be visited from python side.
+
+For example, you can use addexp.a to get the left operand of an Add node.
+
+.. code-block:: python
+
+  x = tvm.var("n")
+  y = x + 2
+  assert(isinstance(y, tvm.expr.Add))
+  assert(y.a == x)
+"""
 # pylint: disable=missing-docstring
 from __future__ import absolute_import as _abs
 from ._ctypes._node import NodeBase, register_node
@@ -75,7 +90,7 @@ class LogicalExpr(Expr):
 
 @register_node("Variable")
 class Var(Expr):
-    """Symbolic variable expression."""
+    """Symbolic variable."""
     pass
 
 @register_node
diff --git a/python/tvm/ir_pass.py b/python/tvm/ir_pass.py
index 3ba8c219a5f9..b43a69700010 100644
--- a/python/tvm/ir_pass.py
+++ b/python/tvm/ir_pass.py
@@ -1 +1,11 @@
-"""Namespace of IR pass functions"""
+"""Namespace of IR pass functions.
+
+This namespace is used for developers. While you do not see any declarations.
+The functions are automatically exported from C++ side via PackedFunc.
+
+Each api is a PackedFunc that can be called in a positional argument manner.
+You can read "include/tvm/pass.h" for the function signature of these functions.
+"""
+from ._ctypes._function import _init_api
+
+_init_api("tvm.ir_pass")
diff --git a/python/tvm/make.py b/python/tvm/make.py
index e874a45b9a2a..50613701cc23 100644
--- a/python/tvm/make.py
+++ b/python/tvm/make.py
@@ -1 +1,11 @@
-"""namespace of IR node builder make function"""
+"""namespace of IR node builder make function
+
+This namespace is used for developers. While you do not see any declarations.
+The functions are automatically exported from C++ side via PackedFunc.
+
+Each api is a PackedFunc that can be called in a positional argument manner.
+You can use make function to build the IR node.
+"""
+from ._ctypes._function import _init_api
+
+_init_api("tvm.make")
diff --git a/python/tvm/module.py b/python/tvm/module.py
index d4c90c6df472..6c9c5f1a365a 100644
--- a/python/tvm/module.py
+++ b/python/tvm/module.py
@@ -1,6 +1,8 @@
 """Runtime module related stuffs"""
 from __future__ import absolute_import as _abs
 from ._ctypes._function import ModuleBase, _init_module_module
+from ._ctypes._function import _init_api
+
 
 class Module(ModuleBase):
     """Module container of all TVM generated functions"""
@@ -28,7 +30,7 @@ def imported_modules(self):
 
         Returns
         ----------
-        modules : list of Modules
+        modules : list of Module
             The module
         """
         nmod = _ImportsSize(self)
@@ -58,7 +60,36 @@ def load(path, fmt=""):
     fmt : str, optional
         The format of the file, if not specified
         it will be inferred from suffix of the file.
+
+    Returns
+    -------
+    module : Module
+        The loaded module
     """
     return _LoadFromFile(path, fmt)
 
+
+def enabled(target):
+    """Whether module runtime is enabled for target
+
+    Parameters
+    ----------
+    target : str
+        The target device type.
+
+    Returns
+    -------
+    enabled : bool
+        Whether runtime is enabled.
+
+    Examples
+    --------
+    The following code checks if gpu is enabled.
+
+    >>> tvm.module.enabled("gpu")
+    """
+    return _Enabled(target)
+
+
+_init_api("tvm.module")
 _init_module_module(Module)
diff --git a/python/tvm/schedule.py b/python/tvm/schedule.py
index 4d22c7b40023..1713b5aa85f2 100644
--- a/python/tvm/schedule.py
+++ b/python/tvm/schedule.py
@@ -3,11 +3,14 @@
 from ._ctypes._node import NodeBase, register_node
 from . import _api_internal
 from . import tensor as _tensor
+from . import expr as _expr
 from . import collections as _collections
+from ._ctypes._function import _init_api
+
 
 @register_node
 class Buffer(NodeBase):
-    """Represent a Buffer in TVM."""
+    """Represent a symbolic buffer in TVM."""
     pass
 
 @register_node
@@ -20,6 +23,29 @@ class Fuse(NodeBase):
     """Fuse operation on axis."""
     pass
 
+@register_node
+class IterVar(NodeBase, _expr.ExprOp):
+    """Represent iteration variable.
+
+    IterVar is normally created by Operation, to represent
+    axis iterations in the computation.
+    It can also created by schedule primitives like :any:`tvm.schedule.Stage.split`.
+
+    See Also
+    --------
+    tvm.thread_axis: Create thread axis IterVar.
+    tvm.reduce_axis: Create reduce axis IterVar.
+    """
+    DataPar = 0
+    ThreadIndex = 1
+    CommReduce = 2
+    Ordered = 3
+    DimInfo = 4
+    Unrolled = 5
+    Vectorized = 6
+    Parallelized = 7
+
+_tensor.iter_var_cls = IterVar
 
 def create_schedule(ops):
     """Create a schedule for list of ops
@@ -343,3 +369,5 @@ def parallel(self, var):
             The iteration to be parallelized.
         """
         _api_internal._StageParallel(self, var)
+
+_init_api("tvm.schedule")
diff --git a/python/tvm/stmt.py b/python/tvm/stmt.py
index 89c923404726..186f4cb31c1c 100644
--- a/python/tvm/stmt.py
+++ b/python/tvm/stmt.py
@@ -1,4 +1,18 @@
-"""Statement classes"""
+"""Statement AST Node in TVM.
+
+User do not need to deal with AST node directly.
+But they can be helpful for developer to do quick proptyping.
+While not displayed in the document and python file.
+Each statement node have subfields that can be visited from python side.
+
+.. code-block:: python
+
+    x = tvm.var("n")
+    a = tvm.var("array", tvm.handle)
+    st = tvm.make.Store(a, x + 1, 1)
+    assert isinstance(st, tvm.stmt.Store)
+    assert(st.buffer_var == a)
+"""
 from __future__ import absolute_import as _abs
 from ._ctypes._node import NodeBase, register_node
 
diff --git a/python/tvm/tensor.py b/python/tvm/tensor.py
index 9c42332b1791..a663490487f1 100644
--- a/python/tvm/tensor.py
+++ b/python/tvm/tensor.py
@@ -1,7 +1,7 @@
-"""Tensor related abstractions"""
+"""Tensor and Computation abstraction objects"""
+# pylint: disable=invalid-name
 from __future__ import absolute_import as _abs
 from ._ctypes._node import NodeBase, SliceBase, register_node, convert_to_node
-from . import collections as _collections
 from . import _api_internal
 from . import make as _make
 from . import expr as _expr
@@ -19,6 +19,7 @@ def __getitem__(self, indices):
             indices = (indices,)
         return TensorSlice(self.tensor, self.indices + indices)
 
+itervar_cls = None
 
 @register_node
 class Tensor(NodeBase):
@@ -30,10 +31,10 @@ def __call__(self, *indices):
         indices = convert_to_node(indices)
         args = []
         for x in indices:
-            if isinstance(x, _collections.IterVar):
-                args.append(x.var)
-            elif isinstance(x, _expr.Expr):
+            if isinstance(x, _expr.Expr):
                 args.append(x)
+            elif isinstance(x, iter_var_cls):
+                args.append(x.var)
             else:
                 raise ValueError("The indices must be expression")
 
@@ -95,20 +96,35 @@ def output(self, index):
         """
         return _api_internal._OpGetOutput(self, index)
 
+
 @register_node
 class PlaceholderOp(Operation):
     """Placeholder operation."""
     pass
 
+
 @register_node
 class ComputeOp(Operation):
     """Compute operation."""
-    pass
+    @property
+    def axis(self):
+        """Represent axis of IterVar, only defined when it is a ComputeOp"""
+        return self.__getattr__("axis")
+
+    @property
+    def reduce_axis(self):
+        """Represent axis of reductions, only defined when it is a ComputeOp"""
+        return self.__getattr__("reduce_axis")
+
 
 @register_node
 class ScanOp(Operation):
     """Scan operation."""
-    pass
+    @property
+    def scan_axis(self):
+        """Represent axis of scan, only defined when it is a ScanOp"""
+        return self.__getattr__("scan_axis")
+
 
 @register_node
 class ExternOp(Operation):
diff --git a/src/api/api_codegen.cc b/src/api/api_codegen.cc
index 9616dccb306a..26f87b352be6 100644
--- a/src/api/api_codegen.cc
+++ b/src/api/api_codegen.cc
@@ -12,7 +12,7 @@
 namespace tvm {
 namespace codegen {
 
-TVM_REGISTER_API(_codegen_build)
+TVM_REGISTER_API(_codegen__Build)
 .set_body([](TVMArgs args, TVMRetValue *ret) {
     if (args[0].IsNodeType<LoweredFunc>()) {
       *ret = Build({args[0]}, args[1]);
@@ -21,7 +21,7 @@ TVM_REGISTER_API(_codegen_build)
     }
   });
 
-TVM_REGISTER_API(_codegen_enabled)
+TVM_REGISTER_API(_codegen__Enabled)
 .set_body([](TVMArgs args, TVMRetValue *ret) {
     *ret = TargetEnabled(args[0]);
   });
diff --git a/src/runtime/module.cc b/src/runtime/module.cc
index 112dca59f50d..6529bb56db5e 100644
--- a/src/runtime/module.cc
+++ b/src/runtime/module.cc
@@ -97,7 +97,7 @@ bool RuntimeEnabled(const std::string& target) {
   return runtime::Registry::Get(load_f_name) != nullptr;
 }
 
-TVM_REGISTER_GLOBAL(_module_enabled)
+TVM_REGISTER_GLOBAL(_module__Enabled)
 .set_body([](TVMArgs args, TVMRetValue *ret) {
     *ret = RuntimeEnabled(args[0]);
     });
diff --git a/tests/python/integration/test_dot.py b/tests/python/integration/test_dot.py
index d74deae1ab7c..2aa3d0ddfe62 100644
--- a/tests/python/integration/test_dot.py
+++ b/tests/python/integration/test_dot.py
@@ -38,7 +38,7 @@ def verify(target):
         if not tvm.codegen.enabled(target):
             print("Target %s is not enabled" % target)
             return
-        f = tvm.codegen.build(fapi, target)
+        f = tvm.codegen.build_module(fapi, target)
         # verify
         ctx = tvm.cpu(0)
         a = tvm.nd.array(np.random.uniform(size=(nn,)).astype(A.dtype), ctx)
diff --git a/tests/python/unittest/test_codegen_device.py b/tests/python/unittest/test_codegen_device.py
index 8ff55acf4e94..90c28a8af363 100644
--- a/tests/python/unittest/test_codegen_device.py
+++ b/tests/python/unittest/test_codegen_device.py
@@ -32,8 +32,8 @@ def check_target(device, host="stackvm"):
         if not tvm.codegen.enabled(device):
             return
         ctx = tvm.gpu(0) if device == "cuda" else tvm.cl(0)
-        mhost = tvm.codegen.build(fsplits[0], host)
-        mdev = tvm.codegen.build(fsplits[1:], device)
+        mhost = tvm.codegen.build_module(fsplits[0], host)
+        mdev = tvm.codegen.build_module(fsplits[1:], device)
         mhost.import_module(mdev)
         code = mdev.get_source()
         f = mhost.entry_func
@@ -54,8 +54,8 @@ def check_module_save(device, host="stackvm"):
             return
         ctx = tvm.gpu(0) if device == "cuda" else tvm.cl(0)
         fmt = "ptx" if device == "cuda" else "cl"
-        mhost = tvm.codegen.build(fsplits[0], host)
-        mdev = tvm.codegen.build(fsplits[1:], device)
+        mhost = tvm.codegen.build_module(fsplits[0], host)
+        mdev = tvm.codegen.build_module(fsplits[1:], device)
         temp = testing.tempdir()
         mpath = temp.relpath("test.%s" % fmt)
         mdev.save(mpath)
diff --git a/tests/python/unittest/test_codegen_vm_basic.py b/tests/python/unittest/test_codegen_vm_basic.py
index e41656e53377..b9d11c3a02c0 100644
--- a/tests/python/unittest/test_codegen_vm_basic.py
+++ b/tests/python/unittest/test_codegen_vm_basic.py
@@ -5,7 +5,7 @@ def run_jit(fapi, check):
     for target in ["llvm", "stackvm"]:
         if not tvm.codegen.enabled(target):
             continue
-        f = tvm.codegen.build(fapi, target)
+        f = tvm.codegen.build_module(fapi, target)
         s = f.get_source()
         check(f)
 
diff --git a/tests/python/unittest/test_lang_basic.py b/tests/python/unittest/test_lang_basic.py
index e9594578acf5..63a109d3c8df 100644
--- a/tests/python/unittest/test_lang_basic.py
+++ b/tests/python/unittest/test_lang_basic.py
@@ -29,6 +29,13 @@ def test_ir():
     stmt = tvm.make.Evaluate(z)
     assert isinstance(stmt, tvm.stmt.Evaluate)
 
+def test_ir2():
+    x = tvm.var("n")
+    a = tvm.var("array", tvm.handle)
+    st = tvm.make.Store(a, x + 1, 1)
+    assert isinstance(st, tvm.stmt.Store)
+    assert(st.buffer_var == a)
+
 def test_let():
     x = tvm.var('x')
     y = tvm.var('y')
diff --git a/tests/python/unittest/test_lang_schedule.py b/tests/python/unittest/test_lang_schedule.py
index de6c5314095b..5fed5a23f750 100644
--- a/tests/python/unittest/test_lang_schedule.py
+++ b/tests/python/unittest/test_lang_schedule.py
@@ -86,8 +86,8 @@ def test_vectorize():
     xo, yo, xi, yi = s[T].tile(T.op.axis[0], T.op.axis[1], x_factor=10, y_factor=5)
     s[T].vectorize(yi)
     s[T].unroll(xi)
-    UNROLL = tvm.collections.IterVar.Unrolled
-    VECTORIZE = tvm.collections.IterVar.Vectorized
+    UNROLL = tvm.schedule.IterVar.Unrolled
+    VECTORIZE = tvm.schedule.IterVar.Vectorized
     assert s[T].iter_var_attrs[xi].iter_type == UNROLL
     assert s[T].iter_var_attrs[yi].iter_type == VECTORIZE
 
diff --git a/tests/python/unittest/test_module_load.py b/tests/python/unittest/test_module_load.py
index 7003b3f92fe4..73de2f6c4fa7 100644
--- a/tests/python/unittest/test_module_load.py
+++ b/tests/python/unittest/test_module_load.py
@@ -36,7 +36,7 @@ def save_object(names):
                            tvm.make.Load(dtype, Ab.data, i) + 1,
                            i + 1))
         fapi = tvm.ir_pass.MakeAPI(stmt, "ramp", [Ab], 0)
-        m = tvm.codegen.build(fapi, "llvm")
+        m = tvm.codegen.build_module(fapi, "llvm")
         for name in names:
             m.save(name)
 
diff --git a/tests/verilog/integration/test_codegen_verilog.py b/tests/verilog/integration/test_codegen_verilog.py
index d8eba1e12188..7003fa936bd1 100644
--- a/tests/verilog/integration/test_codegen_verilog.py
+++ b/tests/verilog/integration/test_codegen_verilog.py
@@ -48,8 +48,8 @@ def check_target(device, host="stackvm"):
         if not tvm.codegen.enabled(device):
             return
         ctx = tvm.vpi(0)
-        mhost = tvm.codegen.build(fsplits[0], host)
-        mdev = tvm.codegen.build(fsplits[1:], device)
+        mhost = tvm.codegen.build_module(fsplits[0], host)
+        mdev = tvm.codegen.build_module(fsplits[1:], device)
         mhost.import_module(mdev)
         code = mdev.get_source()
         f = mhost.entry_func
diff --git a/tutorials/python/README.txt b/tutorials/python/README.txt
index 8c8349b44ef6..66b1b9a20ff1 100644
--- a/tutorials/python/README.txt
+++ b/tutorials/python/README.txt
@@ -1,6 +1,3 @@
-.. _tutorials-index:
-
-TVM Tutorials
-=============
-
-.. _notebook_examples:
+Tutorials
+=========
+These tutorials are tutorial generated by sphinx-gallery.