<p>Traditional models in art and design often focus on classification or style replication, lacking capabilities for both comprehensive extraction and creative combination of design elements. To address this gap, a deep learning framework is proposed that employs a Sheep Flock Optimized Recurrent NeuroNet (SFO-RNN) to enable automatic extraction and dynamic recombination of artistic features such as lines, colors, textures, and spatial layouts. A curated dataset of digital artworks, posters, and illustrations is compiled from public art repositories and open-source design platforms. Preprocessing steps include Adaptive Histogram Equalization (AHE) to enhance contrast and Hampel filtering (HM) for robust outlier removal, ensuring optimal input quality. Convolutional Neural Networks (CNNs) are applied for spatial feature extraction, capturing fine-grained visual patterns. The SFO-RNN, optimized using swarm-inspired behavior, learns contextual relationships between these elements and supports creative combination by intelligently blending and rearranging features to generate novel and aesthetically coherent designs. The framework includes a creative synthesis module that facilitates the generation of unique compositions by recombining elements in varied stylistic and structural configurations. These outputs are integrated into a CAD-based design system, allowing for parameterized manipulation and interactive design exploration. Qualitative evaluations indicate a noticeable improvement in the accuracy of feature extraction and the diversity of creative outputs. The experimental findings show that the suggested SFO-RNN model produced an average Aesthetic Score (AS) of 84.56% and an average Creative Diversity Index (CDI) of 81.07%. It demonstrates enhanced generalization, visual coherence, and aesthetic value when compared to conventional deep learning models.</p>