There are two models:

1. FaceNet: Creates 128D embeddings (Deep Neural Network)
2. SVM Classifier: Input to SVM is 128D embeddings that predicts labels. (Machine Learning)

The pickle file created is for SVM and ‘pickle’ concept is very python specific. To run it on mobile you need to write native c/c++ code for SVM which will save the model output in a UNIX like file with SVM specific format. You can use libsvm C implementation for this purpose. Hope this helps.