In this study, we build hybrid fundamental and empirical models. The empirical models are either conventional linear models or nonlinear models derived using genetic programming (GP). This modeling technique is useful in multiscale modeling for constructing compact representations of experimental data or data generated using complex fundamental models. These representations, which can have tunable fundamental and empirical characteristics, allow us to efficiently leverage information between multiple scales and modeling platforms. We apply this technique to model vapor-liquid equilibrium (VLE) and polymer viscosity. Regarding our particular VLE data, our work indicates that although reasonably accurate models for the excess Gibbs energies can be built using linear models, the models themselves contain a large number of free parameters. The GP-generated models, on the other hand, use nonlinear equations and fewer parameters to capture the dependence of the excess Gibbs energies on temperature and mixture composition. Regarding the polymer viscosity data, good linear models can be easily constructed to capture the dependence of the Williams-Landel-Ferry equation residuals of the viscosity on the molecular characteristics of the polymers. However, the GP-generated nonlinear models are more compact and contain fewer parameters. All of the hybrid models shown here are simple to generate, accurate, and portable, meaning that they call be easily leveraged across a variety of modeling platforms.