The availability of geo-referenced data increased dramatically in recent years, motivating the use of spatial statistics in a variety of research fields, including epidemiology, environmental science, remote sensing, and economics. Combining data measured at both point and areal support can improve parameter estimation and increase prediction accuracy. We propose a new generalized spatial fusion model framework for jointly analyzing point and areal data. Assuming a common latent spatial process, we take a Bayesian hierarchical approach to model both types of data without distributional constraints. The models are implemented with nearest neighbor Gaussian process in Stan modeling language to increase computational efficiency and flexibility. Our simulation study shows that generalized fusion models under this framework model the latent process better than spatial process models. We identify scenarios where fusion models can offer large improvements. We then apply the framework to epidemiological data to identify the spatial risk pattern of respiratory diseases and lung cancer in Canton of Zurich, Switzerland.