Dementia, a condition causing progressive cognitive decline, poses a global health challenge with exponentially increasing yearly patients. The aging population necessitates reliable early detection and classification methods. Hence, early detection of Dementia is critical for the development of effective treatments and interventions. Magnetic Resonance Imaging (MRI) emerges as a valuable tool, allowing visualization of brain structure and detecting structural changes associated with Dementia. It aids in distinguishing Dementia from other forms of Dementia and tracking disease progression, supporting a comprehensive diagnosis when combined with cognitive and neurological assessments. This study comprehensively analyzes various Deep Learning (DL) techniques to address the challenge of early detection of Dementia by proposing a multi-modal cascaded analysis approach for early detection and precise classification. The primary focus is on constructing a robust Dementia Detection and Classification system through a 2-level Deep Learning Classification model. The first level of the model serves as a binary classifier, distinguishing patients with Dementia from those without. This initial screening is crucial for identifying potential cases of Dementia. Whereas, the second level refines the classification, categorizing patients into ‘Very Mild Dementia’, ‘Mild Dementia’, ‘Moderate Dementia’, and ‘Non-Demented’. We evaluated our proposed model and also compared it with current state of the art models such as ResNet-50, DenseNet-121, Inception-V4, Mobilenet-V3 and VGG-16 using a publicly available Kaggle and OASIS dataset.