We present and compare two approaches for the coarse-graining (CG) of models for graphene and carbon nanotubes (CNTs). Such models are required to enable mechanical device simulation on mesoscopic time and length scales hardly reachable by the molecular dynamics method. The first is a heuristic top-down approach while the second performs a rigorous bottom-up CG based upon an atomistic description. Both models belong to the family of dissipative particle dynamics. The top-down model already allows to analyze CNT self assembly and the temperature dependent resonance behavior of resonators. Correct relaxation time-scales required, e.g., for the Q-factor of resonator-devices are hard to adjust in this model. Therefore, a statistical projection-operator based bottom-up approach was investigated. This model allows to reproduce the correct time scales of autocorrelation functions on a CG-level. For correct cross-correlations and hence the correct decay of eigenmodes, further improvements are necessary.