Restructuring of models

All the models are now explicitly described with numpy array. It's a little bit worst from the memory perspective but it's a lot better in terms of running time.

The documentation have been updated and improved.