Towards secure private and trustworthy human-centric embedded machine learning: An emotion-aware facial recognition case study