Can we upload a trained model to classify speech using MFCC's. And then to use it control an arduino via bluetooth.
Perhaps http://appinventor.mit.edu/assets/files/voice-authentication-app-tutorial.pdf shows how to develop a model. Is this part of what you want to do?
You probably want to use the SpeechRecognizer to do the control. Here is a tutorial showing how to use the SR to control activities.
HOW TO: Program the native SpeechRecognizer for Continuous Dictation and to do things.