With an increase in the use of smart devices and data-intensive applications, concerns over privacy have increased since centralized machine learning involves sharing users\' sensitive data. On the other hand, Federated Learning (FL) is a promising technology that allows collaborative learning without sharing actual data. FedSecure is proposed in this study as an effective framework for privacy preserving FL applied in smart applications, which utilizes efficient parameter freezing, differential privacy, and secure aggregation techniques. Specifically, FedFreeze and FedFreeze+ approaches are used in this approach to improve the efficiency of data transmission by 40%. Moreover, in order to ensure effective privacy protection against inference attacks, a distributed differential privacy scheme using additive secret sharing technique is implemented. Experiments carried out on healthcare and personal assistant applications reveal that FedSecure can efficiently protect privacy in a non-IID environment with an accuracy rate of 94.2% at 60% less communication cost.