Eddy diffusivity models are a common method to parameterize turbulent fluxes in the atmospheric sciences community. However, their inability to handle convective boundary layers leads to the addition of a nondiffusive flux component (usually called nonlocal) alongside the original diffusive term (usually called local). Both components are often modeled for convective conditions based on the shape of the eddy diffusivity profile for neutral conditions. This assumption of shape is traditionally employed due to the difficulty of estimating both components based on numerically-simulated turbulent fluxes without any a priori assumptions. In this manuscript we propose a novel method to avoid this issue and estimate both components from numerical simulations without having to assume any a priori shape or scaling for either. Our approach is based on optimizing results from a modeling perspective and taking as much advantage as possible from the diffusive term, thus maximizing the eddy diffusivity. We use our method to diagnostically investigate four different large-eddy simulations spanning different stability regimes, which reveal that nondiffusive fluxes are important even when trying to minimize them. Furthermore, the calculated profiles for both diffusive and nondiffusive fluxes suggest that their shapes change with stability, which is an effect that is not included in most models currently in use. Finally, we use our results to discuss modeling approaches and identify opportunities for improving current models.