Collaborative learning-based training has become popular for developing accurate ML models. However, data owners face constraints regarding privacy and regulatory compliance, limiting their ability to share data. To address this, Federated Learning (FL) has emerged as a solution, enabling collaborative model training without the need to share raw data. Instead, only model parameters are exchanged among participating parties. However, FL faces many challenges, particularly its reliance on a central server, managing data heterogeneity and security challenges of parameter sharing. In this paper, we propose a peer-to-peer federated learning framework that leverages secure multiparty computation (SMC) to overcome these challenges. Our framework eliminates the dependency on a central server, ensuring enhanced security in parameter sharing. We implement our proposed framework for both Independent and Identically Distributed (IID) and non Independent and Identically Distributed (non-IID) data distributions, using FedProx for non-IID data and FedAvg for IID data. We evaluated the performance of our framework using MNIST and FEMNIST datasets, incorporating varying numbers of clients, and conducted a comprehensive analysis. Experimental results demonstrate that our framework performs better in terms of test accuracy and smooth learning than existing centralized FL frameworks.