There is a requirement for localised efficient electricity generation systems that increase the efficiency of power plants and reduce wasted heat from other engineering applications such as cement manufacturing units and brick kilns. Organic Rankine Cycle (ORC) power systems use organic fluids and low-grade heat sources to accomplish this. Lack of preliminary loss models for turbomachines system components, such as turbine and compressor blades, operating in organic fluids prevents the estimation of losses at the preliminary design phase. The design procedure of ORC power system components relies on correlations developed for gas turbines operating with air or steam. The physics-based loss model of Denton is based on assumptions and empirical data to estimate losses. One drawback of this model is the constant dissipation coefficient (C_{d, blade}) value of 0.002. The estimation of C_{d, blade} requires parameters of the boundary layer that are not available at the preliminary design phase. There is thus a requirement for a model to estimate C_{d, blade} as a function of preliminary design parameters such as duty coefficients, volumetric expansion ratio, fluid, operating thermodynamic conditions and blade geometrical parameters. The objective of the present work is to develop a loss model capable of predicting C$_{d, blade}$ based on a database of results from numerical Reynolds-Averaged Navier-Stokes equation simulations for 2D axial stator blades. The developed model is a machine learnt model that is then implemented in an in-house meanline design loss estimation tool, TurboSim, to estimate changes from the current implementation in profile losses and stage efficiency losses. Results from a parametric analysis show that the duty coefficients and working fluid influence on C_{d, blade} the most. The implementation of the machine-learnt model in TurboSim shows that using a constant C_{d, blade} = 0.002 leads to a slight underestimation of the profile loss for all cases except complex molecules operating close to the critical conditions. Although the lack of validation data prevents the estimation of the accuracy of the findings, the results highlight two aspects. First, the need to change the existing assumption of C_{d,blade} = 0.002 for ORC applications. Second, the potential application of data-driven models for loss estimation.