Isomorphic meshes (IMs), which represent all 3D data using a common mesh structure, are one approach to representing 3D data suitable for statistical analysis and deep learning. Existing methods for generating IMs from point clouds include iMG, which suffers from a high computational cost, and DIT-based network, which has limited generalization capability to previously unseen object classes. In this study, we propose a novel method for IM generation that addresses these issues by combining the strengths of iMG and DIT-based network. Our method begins with a global deformation of a template mesh in a manner similar to iMG. Next, both the deformed template and the input point cloud are partitioned into multiple local regions and grouped based on shape similarity. An implicit function network is then applied to each group, using a model trained specifically for that group, following the strategy used in DIT-based network. Finally, all locally deformed regions are integrated to produce the final IM corresponding to the input point cloud. Experiments on 17 object classes demonstrate that the proposed method achieves faster mesh generation than iMG in 15 of them. Furthermore, in 6 classes, it outperforms iMG in both speed and accuracy. We also demonstrate that introducing parallel processing into our framework can further accelerate the mesh generation process.