Hey, we will support more nodes for classification in the upcoming releases that might further abstract decoding. Given the head and post-processing is relatively simple, I think you should be able to export the model to ONNX and compile it through blobconverter directly. Not sure if softmax will be included in the model or not, but this is likely the only operation that would potentially have to run on host.
Let me know if you try it out and run into some issues.
ChrisCoutureDelValle