Inorganic fillers, such as fused silica or organic clay, help tailor/co-optimize the mechanical toughness, thermal conductivity, and moisture diffusivity of polymer mold compounds used to package microelectronic integrated circuits. Despite long history and wide-spread current use, the optimization of filler-infused composites is generally empirical and therefore time-consuming. A physics-based predictive modeling will improve application-specific design of composites that would offer optimum performance and reliability. As an illustrative example, in this paper, we develop a general theory of polymer composites that anticipates the suppression of moisture diffusion as a function of fill-fraction, size-dispersion, shape, and topology of filler nanoparticles. Our results show that the best performance is obtained by incorporation rod-shaped fillers, randomly closed packed at maximum density (~60%). Our numerical results are succinctly captured by the analytical model based on generalized Maxwell Garnett effective medium theory. The analytical model can be used for initial optimization of mold compounds before large-scale numerical modeling is invoked and characterization experiments are designed.