A Paradigm for Potential Model Performance Improvement in Classification and Regression Problems. A Proof of Concept