One rather axiomatic approach is that you call any operator $D$ a derivative that is
Linear:
- $D(\alpha f) = \alpha Df$
- $D(f+g) = Df+Dg$
Obeys Leibniz' Rule:
For "scalars" $\alpha$ and "functions" $f$, $g$. This allows to define derivatives even for functions that operate on discrete sets$^1$ where no concepts like limes or completeness are available. All you need is "addition" and "multiplication", both commutative$^2$.
That is the scalars are $\def\M{\mathcal S} \alpha\in \M=(\M,+,\cdot)$, functions are $f\in \{\M\to\M\}$, and functionals are $D\in \{\{\M\to\M\} \to \{\M\to\M\}\}$
$^1$Not to be confused with discrete derivative.
$^2$Where the latter condition could be dropped. In that case you'll come up with different flavours.