The number of patients diagnosed with brain tumors is on the rise. This highlights the need for a computer-assisted diagnosis system for rapid and accurate detection of brain tumors. This study has developed a lightweight model that is capable of categorizing three different classes of brain tumors. The developed model, BC-detector, uses depthwise separable convolution as the main building block. Depthwise separable convolution achieves comparable effectiveness to conventional convolutional layers while using fewer parameters. Additionally, techniques such as L2 regularization, dropout, and data augmentation were implemented to reduce the risk of overfitting. The BC-detector has obtained a high accuracy rate of more than 94% in categorizing the three classes of brain tumors and an overall accuracy of 97.39%.