We provide a detailed expository report of a new methodology aiming at building a numerical model of the complex pore distribution of porous UO$_2$ ceramics, tunable to real materials, in view of computing their effective thermal behavior. First, based on 2D optical microscopy images, we characterize the material of interest, dedicating a special attention to the porous network because of its major influence on the thermal behavior. Following Meynard et al. (2022), we then propose a simple morphological model combining a Voronoi tessellation and a boolean model, involving a limited number of parameters, from which 3D virtual microstructures (and so 2D cross-sections) can be generated. These parameters are tuned in order to select within our class of models the microstructures that are the most representative of the real ones ; in practice, this optimization process minimizes a cost function based on morphological descriptors computed from the 2D cross-sections. Last, we perform 2D full-field thermal simulations on cross-sections through Representative Volume Elements of both the numerical and the experimental microstructures. We validate our approach by qualitative and quantitative comparisons relative to both global properties and local field statistics.