We report mesoscale simulations of polymer melts and crosslinked polymer networks by using realistic coarse-grained (CG) models that are developed from atomistic simulations of polymer melts. We apply this multiscale strategy to different polymers in order to predict quantitatively some structural and thermomechanical properties such as the melt density, the end-to-end distance, the entanglement mass and the plateau modulus. The temperature dependence of the CG models is investigated through the calculation of the melt specific volumes at different temperatures and the calculation of the isothermal compressibility gives some insight into the pressure transferability of the CG models. We also show that the CG models can be applied successfully to high molecular weight chains. We test the performance of the CG models by calculating directly the plateau modulus of a crosslinked PIB network from mesoscopic simulations under a tensile stress. We compare the value of the plateau modulus with that calculated from the autocorrelation of the stress tensor during equilibrium simulations.