Accreted ice has an abundance of geometric scales, from the size of the gross features such as horns and scallops down to the minute details of individual frozen droplets. A way to avoid the extreme meshing requirements associated with fully resolving the highly complicated iced geometry for computational simulation is proposed. Heat transfer mechanisms are thought to become increasingly important as the ambient temperature approaches the freezing temperature of water from below (often called the “glaze icing” regime). Direct computational resolution of these features is a time consuming process that may potentially increase the size of the computational mesh by orders of magnitude. In addition, an ice shape may have features with extreme concavity or that close in upon themselves. To alleviate these issues, the Discrete Element Roughness Method (DERM) is derived for and implemented in a general-purpose compressible Computational Fluid Dynamics (CFD) code and explored as a way of modeling the sub-resolved roughness scales. This approach is used to capture the relevant physics of ice roughness without prohibitive computational cost. DERM predictions of skin friction and heat transfer compare well with at-plate boundary layer experiments, empirical correlations for pipe flow, and experimental measurements made on a series of ice-roughened airfoils.