Abstract module to support Tensorflow-Lite/Keras models to run inferences on the edge

Since running pre-trained model inferences on the edge targets it's getting more and more common, it would be nice to simplify the integration of models from a couple of wide used frameworks (e.g. Tensorflow-Lite or Keras).

Currently there are some successful paradigms that they do that, but it would be nice if there was just an abstract object (or wrapper class) that was doing all the work in the background and it's simple to use. Also, supporting validation and analysis of the model it would be great, too.

Just dropping this here if anyone sees that's interesting to implement.