This chapter adapts single-metric, simplex-control perceptual quality models, which are learned offline, by combining them into a composite model that defines their tradeoffs. The Application Layer utilizes the composite model at runtime to identify the optimal operating points and achieve the best perceptual quality. To address the exponential complexity involving multiple metrics and controls, we propose a decomposable form that decomposes the original problem into numerous subproblems, each evaluating the perceptual quality of a single metric and a simplex control. We hypothesize that perceptual quality in the composite model depends on the probability of subjects detecting the maximum \(p_{\mathrm {note}}\) of the competing metrics and controls of the subproblems. We prove that the optimal operating points occur when the \(p_{\mathrm {note}}\) s of the competing metrics are equal. This property is significant because it leads to a polynomial-time algorithm for finding the composite quality. We present algorithms for optimizing the composite quality of applications with multi-metric and simplex-control, single-metric and multi-control, and multi-metric and multi-control combinations. In each case, we prove the sufficient conditions for the optimality of the composite perceptual quality. We further illustrate our approach to optimizing the operating points of a two-party VoIP application.