Miscible gas injection is one of the most effective enhanced oil recovery (EOR) techniques. There are several challenges in accurately modeling this process that mostly occur in the near-miscible region. The adjustment of relative permeability for near-miscible processes is the main focus of this work. The dependence of relative permeability on phase identification can lead to significant complications while simulating near-miscible displacements. We present an analysis of how existing methods incorporate compositional dependence in relative permeability functions. The sensitivity of the different methods to the choice of reference points is presented with possible guidelines to limit the modification of the relative permeabilities to physically reasonable values. We distinguish between the objectives of reflecting near miscible behavior and ensuring smooth transitions across phase changes in the existing methods. We highlight an important link that combines the two objectives in a more general framework. We make use of Gibbs free energy as a compositional indicator to honor the generalized framework. The new approach was implemented in the Automatic Differentiation General Purpose Research Simulator (ADGPRS) and tested on a set of near-miscible gas injection problems. We show that including compositional dependencies in the relative permeability near the critical point impacts the simulation results with significant improvements in nonlinear convergence @en