Accurate models of evaporating flows are needed in a range of engineering applications. An important application is to simulate the depressurization and subsequent phase transition of CO2 in pipes. This is relevant for the safe design and operation of CO2 pipelines for large-scale CO2 capture and storage. Experiments of flows with rapid evaporation show significant departure from equilibrium and relaxation models must be applied to describe the process. Lately, a hierarchy of different relaxation models have been developed and studied, and relaxation techniques have been devised for infinite, finite and arbitrary-rate relaxation terms. The relaxation terms applied in these models often require tuning. We explore the potential of applying physics-based relaxation terms in the relaxation models to improve accuracy and avoid tuning. As a first step, this is tested for a model with chemical potential relaxation. The mass relaxation term is modeled using classical nucleation theory (CNT) and a bubble growth term based on non-equilibrium thermodynamics. As CO2 flow in pipelines will often be near the critical point where two-step relaxation methods can reach thermodynamically invalid states, we suggest an implicit relaxation technique in the solution procedure. The full model holds for a general equation of state (EOS), and we apply the Peng-Robinson EOS for the simulation of CO2. We compare the results of this model to experimental data of CO2 depressurizations.