A popular model for NMD in eukaryotes is that it occurs when there is a stop codon prior to the last exon junction complex left over from splicing, so that mRNAs with premature stop codons are destroyed. In general bacterial mRNAs don't undergo splicing, so there wouldn't be any EJCs when the gene is expressed in bacteria. Your question is about taking a bacterial gene and expressing it in eukaryotes and observing NMD. It is possible, but seems unlikely, that the eukaryotic system would find some sequences that resemble eukaryotic splice donor and acceptor sites in the bacterial gene. Have you considered that this is not NMD, but instead some issue related to codon optimization?
I have read a paper somewhere that single exon eukaryotic genes can undergo degradation by NMD. Clearly these genes don't have exon junctions post splicing and so their degradation cannot be mediated by exon junction complexes. In short the 50 bp rule and EJCs is the major model for NMD but much about NMD is not understood. What you proposed is eminently possible. I will try and dig out that particular reference. Are you using a particular eukaryotic expression vector system?
See below from a paper by Neu-Yilik et al., The faux-3'UTR hypothesis could explain what you are seeing. Also there are some genes in which the EJC-50 bp model does not explain which transcripts become substrates for NMD. See a paper by Fang et al about one of the collagen genes.
The below is pasted ===========================================
Mechanism of escape from nonsense-mediated mRNA decay of human β-globin transcripts with nonsense mutations in the first exon
Gabriele Neu-Yilik,1,2 Beate Amthor,1,2 Niels H. Gehring,1,2,4 Sharif Bahri,1,5 Helena Paidassi,1,6 Matthias W. Hentze,2,3 and Andreas E. Kulozik1,2
Recently, splicing-independent and EJC-independent examples of human NMD have also been described. In these cases, very long 3′ UTRs together with an inappropriate spatial arrangement of the ribosome at the stop codon, the 3′ UTR mRNP, and the poly(A) binding protein PABPC1 appear to be able to trigger NMD (Amrani et al. 2004; Buhler et al. 2006; Eberle et al. 2008; Singh et al. 2008). Such an aberrant (or “faux”) 3′ UTR is at least partially defined by the inappropriate distance between the termination codon and PABPC1, which is thought to prevent the termination promoting interaction between eRF3a and PABPC1 (Singh et al. 2008).