Corner separation is a common flow phenomenon within compressors that can significantly affect the compressor performance. The RANS turbulence closures, commonly used in the industrial CFD simulations, often struggle to predict corner separation with reasonable accuracy. In this paper, the results of two RANS-based modelling approaches are presented for the corner separation within a high-loaded Prescribed Velocity Distribution (PVD) compressor cascade. The flow characteristics are studied to facilitate understanding the causes of varying performance of RANS models. It is observed that mixing plays a crucial role in accurately predicting the type, location, and size of flow separation. The source terms that control the turbulence mixing in SA and SST models are identified, based on physical analyses. Both RANS models are modified to better model the mixing process. Based on the modified SST model, an improved RANS-LES blending function has been proposed for a hybrid RANS-LES model. This new blending function ensures reliable shielding of attached boundary layers by the RANS portion of the hybrid model. Finally, to gain further understanding of the endwall flow physics, the turbulence characteristics of the resolved corner separation flow are studied, in terms of the large-scale unsteadiness and loss generation mechanisms.