System-Level Uncertainty Quantification with Multiple Machine Learning Models: A Theoretical Framework