You probably know that Eilenberg and Steenrod introduced axioms to define the concept of a homology theory on an abstract level. Singular homology is one example of such a theory, but there are many other.
Given such a homology theory, a general method to define reduced homology groups is
$$(D) \quad \tilde H_n(X) = \ker (c_* : H_n(X) \to H_n(*))$$
where $c : X\to *$ denotes the unique map to the one-point space $*$. See Tyrone's comment.
However, the standard "textbook approach" for singular homology is to define reduced homology groups $\tilde H_n(X)$ as the homology groups of the augmented chain complex. Doing so, equation $(D)$ becomes a theorem which requires a proof.
Working with the augmented chain complex yields $\tilde H_n(X) = H_n(X)$ for $n > 0$ (both are the same quotient groups), thus trivially $(D)$ is satisfied for $n > 0$ since $H_n(*) = 0$. In dimension $0$ let us observe that $\tilde H_0(X)$ is (in contrast to the kernel definition) not a genuine subgroup of $H_0(X)$, but there is a canonical group monomorphism $\iota : \tilde H_0(X) \to H_0(X)$: The first group is $\ker \epsilon / \text{im} \partial_1$, the second is $C_0(X) / \text{im} \partial_1$ and $\iota$ is induced by the inclusion $\mu : \ker \epsilon \hookrightarrow C_0(X)$. If we understand this point, we may laxly write $\tilde H_0(X) \subset H_0(X)$. However, the precise statement is that $\text{im} \iota = \ker c_*$.
Consider the induced map $c_\# : C_0(X) \to C_0(*)$ on chain complexes. Then
$$\ker c_\# \hookrightarrow C_0(X) \stackrel{c\#}{\to} C_0(*)$$
is trivially exact. But by the definitions of $c_\#$ and $\epsilon$ we have $\ker c_\# = \ker \epsilon$, thus
$$\ker \epsilon \stackrel{\mu}{\to} C_0(X) \stackrel{c\#}{\to} C_0(*)$$
is exact. The induced $c_* : H_0(X) \to H_0(*)$ is given by
$$c_*([\xi]) = [c_\#(\xi)] . $$
Thus $[\xi] \in \ker c_*$ means $[c_\#(\xi)] = 0$. Since the quotient map $C_0(*) \to H_0(*)$ is an isomorphism, the latter is equivalent to $c_\#(\xi) = 0$ and therefore equivalent to $\xi \in \text{im} \mu$ which is the same as $[\xi] = [\mu(\eta)] = \iota([\eta]) \in \text{im} \iota$.
Therefore $\ker c_* = \text{im} \iota$ as desired.