Interpretable to Whom? A Role-based Model for Analyzing Interpretable Machine Learning Systems