On the interplay of Explainability, Privacy and Predictive Performance with Explanation-assisted Model Extraction