Abstract: A majority of the works on Spiking Neural Networks (SNNs) do $not$ deploy and evaluate their network on a neuromorphic hardware. This not only limits the credibility of their claims of energy-efficiency and latency gains, but also discounts the opportunity to appraise neuromorphic technology for real-world computing. Herein, we especially study the technical facets of $deploying$ and $evaluating$ a recently formulated $State$-$Space Model$ based spiking network called “Legendre-SNN” on Loihi-2 neuromorphic hardware. Legendre-SNN is a highly resource-efficient $reservoir$-based univariate Time-Series Classification (TSC) model. This work’s emphasis is not only on its deployment on Loihi-2, but also on leveraging the Loihi-2 embedded Lakemont (LMT) cores for its $non$-$spiking$ reservoir deployment and spike encoding. Since the documentation to program LMT is very limited, researchers often implement their $non$-$spiking$ operations on $less$ power efficient CPUs (than LMTs). Here, we present the technical know-how to program LMT (as part of our reservoir deployment) that can be employed by later works. In our evaluation of Legendre-SNN on Loihi-2 hardware, we pleasantly find that it outperforms a complex LSTM-Conv integrated architecture on 3 of 15 datasets. We also present the $energy$ & $latency$ metrics of Legendre-SNN on Loihi-2, where we find that (for our settings) the reservoir on LMT consumes more than 85% of total energy; consequently, we advocate for a $spiking$ reservoir in Legendre-SNN.