tsai is an excellent package for deep learning models for time series, mostly for time series classification, but some also for forecasting.
https://github.com/timeseriesAI/tsai

It has a unified interface to these, so it should be easy to interface these from sktime. Possibly, even _DelegatedClassifier or _DelegatedForecaster might be used.

As a recipe, I think different classifiers should be different classes; obviously, we also want to credit the authors of the individual classifiers, with their GitHub names, in the "authors" tag - these can be gleaned from the commit history.

FYI @oguiza, it would be great to collaborate on this! Ref issue on tsai: timeseriesAI/tsai#931