In spatial statistics, data are often collected at different spatial resolutions. Often, it is of interest to (a) carry out multivariate analysis when variables are sampled at different locations, (b) model data collected at misaligned areas, or (c) unravel common latent factors by jointly modelling point and areal data. In this paper, we establish a linkage between the generalized spatial fusion model framework and the various change‐of‐support problems, and we outline how the framework can be adapted in these situations. Moreover, we propose an efficient fusion model implementation by exploiting advantages of nearest neighbour Gaussian process and the Stan modelling language. Our simulation shows that the computational efficiency is several times higher in the new implementation compared with original implementation. We illustrate the performance gain in practice using a case study, which models daily precipitation in Switzerland based on rain gauge and radar data.