Constrained devices, such as smart sensors, wearable devices, and Internet of Things nodes, are increasingly prevalent in society and rely on secure communications to function properly. These devices often operate autonomously, exchanging sensitive data or commands over short distances, such as within a room, house, or warehouse. In this context, continuous-variable quantum key distribution (CV-QKD) offers the highest secure key rate and the greatest versatility for integration into existing infrastructure. A key challenge in this setting, where devices have limited storage and processing capacity, is obtaining a realistic and tight estimate of the CV-QKD secure key rate within a composable security framework, with error correction (EC) consuming most of the storage and computational power. To address this, we focus on low-density parity-check (LDPC) codes with non-binary alphabets, which optimise mutual information and are particularly suited for short-distance communications. We develop a security framework to derive finite-size secret keys near the optimal EC leakage limit and model the related memory requirements for the encoding process in one-way error correction. This analysis facilitates the practical deployment of CV-QKD, particularly in constrained devices with limited storage and computational resources.