Localization, Deepening, and other bells and whistles

## Issue
I have recently come across the excellent article of https://www.maths.usyd.edu.au/u/gottwald/preprints/DeepRFM.pdf

Though focused on the eventual goal of autonomous forecasting of dynamical systems, they introduce some developments that are well worth exploration
```[tasklist]
- [ ] A method of deepening the model that appears effective.
- [ ] A method of Spatial localization, that could be used if the forward map has a local structure over the inputs.
- [ ] A new method/viewpoint of training RFM to pick features that express high nonlinearity over necessarily optimization
```