This thesis describes the design, implementation, and evaluation of efficient algorithms for self-gravity simulations in astronomical agglomerates. Due to the intrinsic complexity of modeling interactions between particles, agglomerate are studied using computational simulations. Self-gravity affects every particle in agglomerates, which can be composed of millions of particles. So, to perform a realistic simulation is computationally expensive. This thesis presents three parallel multithreading algorithms for self-gravity calculation, including a method that updates the occupied cells on an underlying grid and a variation of the Barnes & Hut method that partitions and arranges the simulation space in both an octal and a binary tree to speed up long range forces calculation. The goal of the algorithms is to make efficient use of the underlying grid that maps the simulated environment. The three methods were evaluated and compared over two scenarios: two agglomerates orbiting each other and a collapsing cube. The experimental evaluation comprises the performance analysis of the two scenarios using the two methods, including a comparison of the results obtained and the analysis of the numerical accuracy by the study of the conservation of the center of mass and angular momentum. Both scenarios were evaluated scaling the number of computational resources to simulate instances with different number of particles. Results show that the proposed octal tree Barnes & Hut method allows improving the performance of the self-gravity calculation up to 100 with respect to the occupied cell method. This way, efficient simulations are performed for the largest problem instance including 2,097,152 particles. The proposed algorithms are efficient and accurate methods for self-gravity simulations in astronomical agglomerates.