feasibility of providing as a couple of simple tensorflow models

I am curious about whether this could be provided as a simple tensorflow model. I.e. something where one could just:

```python
lyra_encoder = keras.models.load_model('path_to_lyra_encoder_model')
lyra_decoder = keras.models.load_model('path_to_lyra_decoder_model')
```

and from there one would be able to simply

```python
encoded_sound_frame = lyra_encoder.predict(sound_frame)
decode_sound_frame = lyra_decoder.predict(encoded_frame)
```

Releasing lyra in such a way would make it so much easier to use on a wide range of environments, from microcontrollers to gpus.

Is there any fundamental reason for such a way to release to not work, or "is it just not how it is done as of today"?