Brain tumors are among the most fatal diseases. If brain tumors are diagnosed in the early stage, then an effective treatment procedure can be planned otherwise the chances of survival are very few. Manual diagnosis of brain tumor is not possible as the location of brain tumor is within the skull. Thus, for the diagnosis purpose, an accurate and non-invasive method is needed. For correct diagnosis of a brain tumor, a machine with high accuracy is desired. Computerized diagnosis systems are proving to be the most efficient in detection and classification tasks. Advancements in deep learning have revolutionized the medical imaging related research. Deep learning methods have outperformed the other traditional methods. Continuous efforts are being made by the researchers to enhance the detection accuracy. This work proposes an architecture for enhanced brain tumor detection accuracy. Ours proposed architecture has an accuracy of 98.85% on the test dataset, which is better than the other methods stated in the literature survey. Our proposed architecture has a very low complexity as compared with other architectures which saves the time as well as the computational efforts, and hence saves energy.